Open-source image generation API. Every picture starts blank.
Windpaint is an API for generating images with open-source models, with storage for everything you make. Video and workflows are coming soon.
Move to paint. Click the paper to throw a picture.Click the paper to throw a picture.Tap the paper to throw a picture.

Each of these started as a sentence.
Real outputs, not retouched, with the prompt and seed that made them. Same seed, same request, same picture. Click a painting to see all of it.







Write a sentence. Get a picture.
Z-Image Turbo makes an image in 8 steps: $0.004 at 1024 px. Ten aspect ratios, up to four images per request, PNG, JPEG or WebP. Presets wrap your prompt in a Windpaint template when you want a house style.
FLUX.2 [klein] 4B is live too, at $0.005 an image. Coming soon: Wan 2.2 image-to-video, Z-Image, Qwen-Image-2512, FLUX.1 [schnell], HiDream-O1-Image, Stable Diffusion XL, Qwen-Image-Edit-2511, Wan 2.2 text-to-video, Wan 2.2 TI2V 5B, LongCat-Video and HunyuanVideo 1.5.

Product shot on seamlessA frosted amber glass perfume bottle

Cinematic stillA lighthouse keeper reading by lamplight during a storm

Isometric illustrationA floating island with a windmill and a vegetable garden

Anime key artA young courier flying a small propeller plane over a coastal town

Editorial portraitAn elderly jazz trumpeter in a linen shirt

Flat vector illustrationA fox reading under a tree at night

Overhead food photoA summer peach and burrata salad

Nothing you make gets lost.
Every upload and output is stored as an asset in your project, with an id. Pass the id straight into the next job instead of downloading and re-uploading. List them, download them, mint a signed URL for an <img> or <video> tag, delete them, or set how long they stay.
- id goes straight into the next job
- signed URL valid for 15 minutes
- 7 days of storage free, then 2 credits per GiB per 30 days
One idea in. A still, a clip and a cover out.
Workflows chain generation steps and free processing steps, like resize and extract frame, behind one request. Text to Clip will take an idea and return a cinematic still, a 5 second 480p clip from Wan 2.2 and a cover frame, each saved as an asset, for 4.08 credits ($0.204). It’s coming soon, along with video generation.
Runs are estimated before they start and followed by polling, a signed webhook or an event stream. You only pay for the steps that finish.

Still

Clip

Cover frame
Your agent can paint too.
A hosted MCP server gives Claude Code, Cursor, Codex and any MCP client that can send a header 20 tools: catalog, estimate, upload, generate, wait, download and workflows. Agent skills come packaged as a plugin.

Pay for finished work.
Prepaid credits at $0.05 each. You only pay for jobs that complete.

- $0.004
- an image. Z-Image Turbo at 1024 px.
- $0.005
- an image. FLUX.2 [klein] 4B at 1024 px.
- $0.20
- a clip, coming soon. Wan 2.2, 5 seconds at 480p.
- 200
- free credits when your early-access account is approved, valid for 30 days.
We don’t train on your prompts, inputs or outputs. The live models are Apache-2.0, so what you make with them can be used commercially. Prompts and image outputs are moderated, and outputs are kept for 7 days unless you say otherwise.

Start yours.
Request early access. We’ll email you when your account is approved, with 200 free credits to spend in the first 30 days.